Transcript here

IEEE Fort Worth PES Chapter
Novel Analytics Solutions to Several
Significant Inconsistencies in Smart Grid
Dr. Yu Meng, Data Scientist - Oncor Electric Delivery
November 15, 2016
Session details


Title: Novel Analytics Solutions to Several Significant Inconsistencies in Smart Grid
Description: Power grids may be the largest and most complex M2M network systems on the globe. IoT
technologies are changing the operation and management paradigm of the grid. In addition to SCADA,
Distribution Management Systems, Outage Management Systems (OMS), and Advanced Metering
Infrastructure (AMI) support new generation of grid monitoring, recording and control, improving grid
resilience, fault management, and accurate measurement. The integration of data, statistics models and data
mining techniques offer vast opportunities to increase operational efficiency and enhance the customer
experience.
Join this session to learn about innovative solutions to several inconsistencies occurring in the Smart Grids
and the results Oncor has achieved including faster, more labor-efficient, and more accurate in
SAIDI/CAIDI calculations, automatic OMS record corrections, and distribution model connectivity error
detection. Identification of characteristic asset data-sets from Smart Grid IoT generated data-sets, data
mining algorithms, and Oncor’s analytic platform pilot will be discussed.

Date/Time: Tuesday, November 15th, 12:00-1:00pm
2
INTRODUCTION
• The 6th largest utility in the United States 3,310,530 accounts
• Serving over 10 million consumers (more
than 1/3 of the state’s population)
• 105,168 miles of distribution lines
• 120,000 miles of transmission lines
• 53,486 miles2 service territory
• 4 million network nodes
• 890,000 distribution transformers
• 1370 substations and 3176 feeders
• 20 PMUs
• Highest electric demand growth region in
the United States
At ONCOR, Smart Grid Includes…
Communications
Radio Frequency (RF) Mesh,
Cellular, Fiber, Satellite,
Microwave, Pager
Distribution Automation
Distribution Management Sys
Distributed Intelligence
Remote Control
Fault Indication
Voltage / Current Sensing
Capacitor Control
Mobile Workforce Management
Outage Management System
Distribution SCADA
Distribution Network Applications
Advanced Metering System
Remote Reading
Remote Connect / Disconnect
Outage Detection
Outage Management System Interface
The Evolution of DMS
Upgraded to
InService 8.2
Installed
InService 8.1
Distribution
Upgraded to
SCADA
InService 9.2
Completed
iFactor Map
Enabled
Upgraded to
InService 8.3.1
Began
OMS
Rollout
2008
2009
Began MWM
Rollout
Completed
MWM Rollout
2010
2011
2012
Completed
OMS Rollout
AMS Interface
Installed Siemens
Turned On
Spectrum
2014
Installed Siemens
DNA Enabled
2016
Upgraded to
InService 9.3
Advanced Metering System
2004
• 3.2 million meters
• 35,000 remote read meters (Phone, Cellular, ERT, PLC)
2004 - 2007
• 500,000 PLC
• 100,000 BPL
2008
• Surcharge approved by the
PUC
• Start AMS deployment
2012
• AMS Deployment 2008 – 2012
• 3.26 million AMS meters
2012- present
Steady grow to 3.34 million
ANALYTICS PLATFORM
EDW
BI/Visualization
4V’s of Big Data
Volume - Scale of Data: SCADA, AMS/AMI, PMU
Velocity - Streaming of Data: SCADA, PMU
Variety - Forms of Data: SCADA, OMS, AMI/AMS, CIS, DIS, Asset DB…
Veracity - Uncertainty of Data: OMS, AMS Interval data, AMS Event log…
PROBLEM
Distribution Outage Event Records
IVR
SMS
Web
Call
Meter
Operator Creation
Level of outages
•
•
•
•
Service level
Transformer level
Feeder level
Substation level
Event Creation
OMS Event Data
Event number
Create Time
Close Time
Cause
Step of Restoration
No. of Customers
Event Closed
Applications
•
•
•
•
PUC Report
SAIDI
SAIFI
Company Record
Problems?
•
•
Connectivity Errors
Human Inaccuracy
How AMS Can Help?
Meter Event Log Data
Meter Interval Data
•
•
•
•
Log only – Upon request
Alerts – Every 4 hours
Alarms – Pushed immediately
Meter number
Event type
Outage time
Restore time
•
Every 15 min, or 96
readings/day
5 day gap retrieval
Meter number
KWH
Average Voltage *
Power Status Flag: PART,
SKIP, GOOD
How AMS data can help?
• Instrumental measures
• Finest measurements
OUR SOLUTION
Use Both OMS and AMS data
OMS Event Data
Event number
Create Time
Close Time
Cause
Step of Restoration
No. of Customers
Connectivity Model
Meter-Event Data
Meter Number
Event number
Create Time
Close Time
Cause
Steps of Restoration
Meter Event Log Data
•
•
•
Log only – Upon request
Alerts – Every 4 hours
Alarms – Pushed immediately
Meter number
Event type
Outage time
Restore time
The Reality Is…
The Problem Becomes……
It is a FUZZY, MANY-TO-MANY,
MATCHING problem.
• Rule based programming
• Data mining
Use AMS Only? Listen to what Bayes Theorem says…
𝑃 𝐴𝐵 =
𝑃 𝐵 𝐴 ∗ 𝑃(𝐴)
𝑃(𝐵)
Legend
𝑃 𝐵 = 𝑃 𝐵 𝐴 ∗ 𝑃 𝐴 + 𝑃 𝐵|_𝐴 ∗ 𝑃(_𝐴)
P(B/A)= 99.5%
P(A) = 100/500,000
Outage
P(_B/A)= 0.5%
Power
P(_A) = 499,900/500,000
Normal
A: Event - Power outage
B: Emission – Detected outage
P(A): Possibility of outage for a meter.
P(B): Possibility a meter reports an outage.
P(B|A): Accuracy of a meter.
Positive
𝑃 𝐴𝐵 =
100
500,000
99.5%∗
100
499,000
+0.5%∗
500,000
500,000
99.5%∗
= 4%
Negative
P(B/_A)= 0.5%
Positive
P(_B/_A)= 99.5%
Negative
Given an outage event, it has at
least 4% of chance to be a real
outage.
Use AMS Only? Listen to what Bayes Theorem says…
Use AMS only – to achieve P(A|B) = 95%
P(B|A) =?
𝑃(𝐴|𝐵)∗100 500,000
𝑃(𝐴|B)∗100 500,000+499,000/500,000∗(1−𝑃(𝐴|𝐵))
= 95%
P(B|A) ~ 99.999%
Use AMS only -
Use both OMS and AMS -
Event: Power Outage P(A) = 100/500,000
Emission: Power Failure Flag P(B) =
500/500,000
Assume P(B|A) = 99.5%
Then P(A|B) = 99.5%/5 = 20%
Event: Power Outage
P(A) = 85%
Emission: Power Failure Flag P(B) = 83%
Assume P(B|A) = 99.5%
Then P(A|B) = 99.6%
Is the Data Exhaustive?
3,300,000
30,000
OMS
OMSAMS
Gap – Rolling out Power Failure Flags
and 15 Min Average Voltage
The Process is
Merge
OMS
Matching & Scoring
AMS
EVENTS
AMS
CUMSUPTION
ETL
ETL
ETL
Clustering
EDW
De-Normal
Master Sheet
Agg Sheet
Flagging
DIS
Aggregate
Data Sources
EDW
Analytics
Presentation
Matching and Scoring
Drill down to meter level
Start
Get OMS Event Records
Get DIS Connectivity Model
OMS <= function getEvent-MeterRecords(OMS events, DIS Connectivity Model)
AMS <= function getAMS-Outage/RestoreEvent()
For Each meter X: OMS
Get a list of OMS records for X
Get a list of AMS records for X
For Each Record
Match and score the similarity
End For Loop
End For Loop
End
Matching and Scoring
OMS
AMS
Time
What the Score tells us?
Score > 0
Score = 0
Score = -1
Cluster Analysis
Duration
Duration
Roll up to event level
Time Off
Time Off
Cluster Analysis
Duration
Mapping meters back to the step of restorations
Time Off
Clustering Analysis
•
•
•
•
Density: No of points within a specified radius R (EPS)
MinPts: a minimum density threshold in EPS
Core Point: A point having more than minpts within EPS
Border Point: A point having fewer neighbors than
minpts within EPS, but in the neighborhood of a core
point
• Noise Point: Any point that is not a core point or a
border point.
Density
EPS
MinPts = 4
Clustering Analysis
Noise Point
Border Point
Core Point
EPS
MinPts = 4
Density Based Clustering
Applications – DIS Connectivity Model Validation
OMS
AMS
AVG VOLT
GPS
VALID
VALID
VALID
INVESTIGAE/
EXCLUDE
Applications - SAIDI
VALIDATE
OUTAGES
USE OMS TO
FILTER EVENT
TYPES
CALCULATE WITH
IMPROVED
DATASET
Applications - Edit Correction
OUTAGE/RESTORE
TIMES
CUSTOMER
COUNT
STEPS OF
RESTORATION
CONCLUSION
LOCATE YOUR
ASSET DATA
MANAGED
DATA AND
AVAILABILITY
ANALYTICS
PLATFORM
AND
PARADIGM
DATA MINING
AND STATISTIC
ANALYSIS
BUSINESS
APPLICATIONS